[
https://issues.apache.org/jira/browse/HBASE-20875?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16546018#comment-16546018
]
stack commented on HBASE-20875:
-------------------------------
bq. In the jitviewer also noticed that in the hot path (in read path in this
case) there are instances where it says 'callee too big' because of which
inlining does not happen.
Yeah. If can make it inline by making stuff smaller and/or dumbing down the
options/types, usually goes faster.
> MemStoreLABImp::copyIntoCell uses 7% CPU when writing
> -----------------------------------------------------
>
> Key: HBASE-20875
> URL: https://issues.apache.org/jira/browse/HBASE-20875
> Project: HBase
> Issue Type: Sub-task
> Components: Performance
> Affects Versions: 2.0.1
> Reporter: stack
> Assignee: stack
> Priority: Major
> Fix For: 2.0.2
>
> Attachments:
> 0001-HBASE-20875-MemStoreLABImp-copyIntoCell-uses-7-CPU-w.patch,
> 2.0707.baseline.91935.cpu.svg, 2.0711.patched.145414.cpu.svg,
> HBASE-20875.master.001.patch, HBASE-20875.master.002.patch, Screen Shot
> 2018-07-11 at 9.52.46 PM.png
>
>
> Looks like this with a PE random write loading:
> {code}
> ./hbase/bin/hbase --config ~/conf_hbase
> org.apache.hadoop.hbase.PerformanceEvaluation --nomapred --presplit=40
> --size=30 --columns=10 --valueSize=100 randomWrite 200
> {code}
> ... against a single server.
> {code}
> 12.47% perf-91935.map
> [.] Lorg/apache/hadoop/hbase/BBKVComparator;::compare
> 10.42% libjvm.so
> [.]
> ParNewGeneration::copy_to_survivor_space_avoiding_promotion_undo(ParScanThreadState*,
> oopDesc*, unsigned long, markOopDesc*)
> 6.78% perf-91935.map
> [.]
> Lorg/apache/hadoop/hbase/regionserver/MemStoreLABImpl;::copyCellInto
> ....
> {code}
> These are top CPU consumers using perf-map-agent ./bin/perf-java-top...
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
